Let’s start by looking at the entire images. The lesion is only a very small part of it.
I am not exactly sure what to do here, since we do not have two measures for each pixel. Also, the size/location of lesion would change across scans and site. There is no one-to-one relationship here.
If the goal is to generate image, then it would not be a problem. But if the goal is to generate WMF and QSM, we would at least need to know how these values are calculated.
I think we would need to focust on the lesion part (is that correct?) and also that would reduce the data by quite a bit. So I hope to start there.
I think I can start by doing things with the whole image. At least there would be one-to-one relationship across the scans at the same site.